Statistical analysis

To test the effects of fertilization, mowing frequency and sward composition in time series data (NO3

-, DON leaching and N2O emissions) we used linear mixed effects models (LME) with treatments factors and row / column of the Latin rectangle design as fixed effects.

Sampling date and spatial replication were included as random effects. The LME model included 1) a variance function that allows different variances of the response variable for the fixed effects (Zuur et al. 2009), 2) a first-order temporal autoregressive process that assumes the correlation between measurements decreases with increasing time difference, or both if these improve the relative goodness of model fit based on the Akaike Information Criterion (AIC) (Crawley 2007). Pairwise comparisons (T-Test with Holmes-Correction) were used as post-hoc tests. To test treatment effects on gross N mineralization, NH₄⁺ immobilization, total average N losses, and N retention efficiency we used analysis of variance (ANOVA) followed by Tukey’s HSD post-hoc test. In all tests, if residual plots revealed non-normal distribution or non-homogeneity of variance we used either logarithmic or square root transformation (after adding a constant value if the dataset included negative values) and analyses were repeated. Non-significant interactions as well as row and column were removed stepwise from the statistical models if this improved AIC (Crawley 2007).

Correlations between N retention efficiency, soil microbial C and N and biomass yield were tested with Spearman’s rank correlation. A significance level of α = 0.05 was used throughout unless stated otherwise. All statistical analyses were performed using the R version 2.11.1 (R Development Core Team 2009).

3.3 Results

3.3.1 Gross N transformation rates and microbial biomass

Rates of gross N mineralization ranged from 71 to 1440 mg N m^-2 d^-1 with an overall mean of 606 (± 65 SE) mg N m^-2 d^-1. Neither fertilization nor sward composition affected gross N mineralization (Table 8). In contrast, gross nitrification rates were higher on the fertilized plots, while sward composition did not affect gross nitrification. NH4+

immobilization (Table 8) varied between 143 and 2356 mg N m^-2 d^-1 with an overall mean of 753 (± 100 SE) mg N m^-2 d^-1 and was lower on the fertilized plots (marginally significant). While net NH4+

immobilization (i.e. NH4+

immobilization rates > gross N mineralization) occurred on the

58

unfertilized plots, on the fertilized plots NH₄⁺ immobilization was 90% of gross N mineralization with the exception of dicot-enhanced swards which displayed net NH4+

immobilization on both fertilized and unfertilized plots. Microbial C was not affected by any of the treatments, but fertilization resulted in marginally lower microbial N contents. A marginally significant interaction between fertilization and sward composition suggested that the decrease in microbial N caused by fertilization was most pronounced at the monocot-enhanced swards. A marginally significant interaction between fertilization and sward composition could also be shown for microbial C:N ratios which were highest at the fertilized monocot-enhanced swards.

59

Table 8: Gross N mineralization rates, gross nitrification rates, NH4+

immobilization rates, microbial C, microbial N and microbial C:N ratios of a grassland under different management practices in the Solling uplands, Germany (mean values + SE).

Main factors

Mean values with different letter indicate significant (P ≦ 0.05) or marginally significant (P ≦ 0.1) differences within main factors (two-way ANOVA).

Average measured bulk density of 0.79 g cm^-3 and a depth from 0.00-0.05 m was used to convert determined dry mass based rates to area based rates.

60

3.3.2 Water balance and N losses

During the experiment, total annual precipitation was 1001 mm in 2009 and 1083 mm in 2010. Modelled annual evapotranspiration was 507 mm in 2009 and 484 mm in 2010.

Modelled drainage flux was 441 mm in 2009 and 609 mm in 2010. At all three depths measured matrix potential was correlated with modelled matrix potential (0.2 m: Spearman’s correlation coefficient = 0.66, P = 0.001, n = 20; 0.5 m: Spearman’s correlation coefficient = 0.58, P = 0.007, n = 20; 0.9 m: Spearman’s correlation coefficient = 0.745, P = 0.018, n = 10).

In both years, modelled drainage flux was negligible in the summer months and strongly increased to values of more than 3 mm d^-1 in autumn. In 2010, the increase in drainage flux occurred about one month earlier compared to 2009 (data not shown). Parallel to drainage flux, NO3

leaching losses were also negligible in the summer months (Fig. 4a, b, c). In fertilized plots, NO3

leaching strongly increased during autumn. Compared to 2009, when the main peak in NO3

leaching appeared only in early 2010, the increase in NO3

leaching following fertilization in 2010 was more pronounced and much earlier. In 2009, fertilization was the only factor that influenced NO3

leaching (P = 0.026). In 2010, NO3

leaching. An interaction (P = 0.011) between these factors showed that the increase of NO3

leaching losses caused by fertilization was only significant for plots cut once per year while plots cut three times per year were not affected. Sward composition did not affect NO₃^- leaching.

Leaching of DON ranged between 0 to 1.9 mg N m^-2 d^-1 with an overall mean of 0.26 (± 0.01 SE) mg N m^-2 d^-1 in 2009. None of the treatment factors influenced DON leaching (Fig. 4d, e, f). In 2010, DON leaching increased to an overall mean of 0.6 (± 0.1 SE) mg N m

-2 d^-1 ranging from 0 to 13.06 mg N m^-2 d^-1. We detected an interaction between the factors fertilization and mowing frequency (P = 0.006): on the plots cut once per year, fertilization significantly increased DON leaching while plots cut three times per year were not affected.

No effect of sward composition on DON leaching could be detected.

61

Figure 4: NO₃^- and DON leaching rates (± SE, n=6) at 0.5-0.6 m mineral soil of a grassland under different management practices in the Solling uplands, Germany: one mowing per year without fertilization (○), one mowing per year with fertilization (●), three mowings per year without fertilization (□), three mowings per year with fertilization (■).

62

In 2009, N₂O emission rates varied between -3.87 to 8.07 mg N₂O-N m^-2 d^-1 with an overall mean of 0.32 (± 0.04 SE) mg N2O-N m^-2 d^-1. Both fertilization (P = 0.000) and mowing frequency (P = 0.031) influenced N2O emissions in 2009 (Fig. 5a, b, c). Furthermore there was an interaction between these factors (P = 0.009). Fertilized plots showed an increase in N2O fluxes, especially following the second fertilizer application in May. However this increase was only significant for the plots cut once per year. Unfertilized plots only showed a marginal increase in N2O fluxes during the summer months. In 2010, the significant effect of fertilization (P = 0.000) could also be observed (Fig. 5d, e, f) with an overall mean of 0.45 (± 0.06 SE) mg N2O-N m^-2 d^-1 (range between -1.84 to 13.58 mg N2O-N m^-2 d^-1). Again, the increase of N2O emissions occurred after the second fertilization in the beginning of July and the impact of fertilization tended to be stronger on the plots cut once per year, however, unlike 2009, this interaction was not significant (P = 0.108). In both years, N2O uptake predominantly occurred on unfertilized plots. There was no impact of sward composition on N2O emissions.

63

Figure 5: N₂O emissions (± SE, n=6) of a grassland under different management practices in the Solling uplands, Germany: one mowing per year without fertilization (○), one mowing per year with fertilization (●), three mowings per year without fertilization (□), three mowings per year with fertilization (■).

64